exponentially-fitted methods applied to fourth-order boundary ...mvdaele/voordrachten/kos2008.pdf•...

Introduction Fourth-order boundary value problems Numerical examples Conclusions

Exponentially-fitted methods applied tofourth-order boundary value problems

M. Van Daele, D. Hollevoet and G. Vanden Berghe

Department of Applied Mathematics and Computer Science

Sixth International Conference of Numerical Analysis andApplied Mathematics, Kos, 2008

Outline

IntroductionAn exampleExponential fitting

Fourth-order boundary value problemsThe problemExponentially-fitted methodsParameter selection

Numerical examplesFirst exampleSecond example

Conclusions

Introduction

In the past 15 years, our research group has constructedmodified versions of well-known

• linear multistep methods

• Runge-Kutta methods

Aim : build methods which perform very good when the solutionhas a known exponential of trigonometric behaviour.

Linear multistep methods

A well known method to solve

y ′′ = f (y) y(a) = ya y ′(a) = y ′a

is the Numerov method (order 4)

yn+1 − 2 yn + yn−1 =1

12h2 (f (yn−1) + 10 f (yn) + f (yn+1))

Construction :impose L[z(t); h] = 0 for z(t) ∈ S = {1, t , t2, t3, t4} where

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

Linear multistep methods

A well known method to solve

y ′′ = f (y) y(a) = ya y ′(a) = y ′a

is the Numerov method (order 4)

yn+1 − 2 yn + yn−1 =1

12h2 (f (yn−1) + 10 f (yn) + f (yn+1))

Construction :impose L[z(t); h] = 0 for z(t) ∈ S = {1, t , t2, t3, t4} where

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

A model problem

Consider the initial value problem

y ′′ + ω2 y = g(y) y(a) = ya y(a) = y ′a .

If |g(y)| ≪ |ω2 y | then

y(t) ≈ α cos(ω t + φ)

To mimic this oscillatory behaviour, one could replacepolynomials by trigonometric (in the complex case :exponential) functions.

A model problem

y ′′ + ω2 y = g(y) y(a) = ya y(a) = y ′a .

A model problem

y ′′ + ω2 y = g(y) y(a) = ya y(a) = y ′a .

EF Numerov methodConstruction : impose L[z(t); h] = 0 for z(t) ∈ S with

S = {1, t , t2, sin(ω t), cos(ω t)}

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

yn+1 − 2 yn + yn−1 = h2 (λf (yn−1) + (1 − 2 λ) f (yn) + λ f (yn+1))

4 sin2 θ2

−1θ2 θ := ω h

θ2 +1

6048θ4 + . . .

S = {1, t , t2, sin(ω t), cos(ω t)}

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

4 sin2 θ2

−1θ2 θ := ω h

θ2 +1

6048θ4 + . . .

S = {1, t , t2, sin(ω t), cos(ω t)}

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

4 sin2 θ2

−1θ2 θ := ω h

θ2 +1

6048θ4 + . . .

S = {1, t , t2, sin(ω t), cos(ω t)}

L[z(t); h] := z(t + h) + α0 z(t) + α−1 z(t − h)

−h2 (

β1 z ′′(t + h) + β0 z ′′(t) + β−1 z ′′(t − h))

4 sin2 θ2

−1θ2 θ := ω h

θ2 +1

6048θ4 + . . .

EF methodsGeneralisation : to determine the coefficients of a method, weimpose conditions on a linear functional. These conditions arerelated to the fitting space S which contains

• polynomials :{tq|q = 0, . . . , K}

• exponential or trigonometric functions, multiplied withpowers of t :

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

{tq cos(ωt), tq sin(ωt)|q = 0, . . . , P}

EF method can be characterized by the couple (K , P)Classical method : P = −1number of basis functions : M= 2 P + K + 3

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

{tq exp(±µt)|q = 0, . . . , P}

or, with ω = i µ,

Examples

M = 2 P + K + 3

(K , P)

M = 2 M = 4 M = 6 M = 8 M = 10(1,−1) (3,−1) (5,−1) (7,−1) (9,−1)(−1, 1) (1, 0) (3, 0) (5, 0) (7, 0)

(−1, 1) (1, 1) (3, 1) (5, 1)(−1, 2) (1, 2) (3, 2)

(−1, 3) (1, 3)(−1, 4)

(1, 2) =⇒ S ={

1, t , exp(±µt), t exp(±µt), t2 exp(±µt)}

Examples

M = 2 P + K + 3

(K , P)

M = 2 M = 4 M = 6 M = 8 M = 10(1,−1) (3,−1) (5,−1) (7,−1) (9,−1)(−1, 1) (1, 0) (3, 0) (5, 0) (7, 0)

(−1, 1) (1, 1) (3, 1) (5, 1)(−1, 2) (1, 2) (3, 2)

(−1, 3) (1, 3)(−1, 4)

(1, 2) =⇒ S ={

1, t , exp(±µt), t exp(±µt), t2 exp(±µt)}

Exponential FittingL. Ixaru and G. Vanden BergheExponential fittingKluwer Academic Publishers, Dordrecht, 2004

η−1(Z ) =

cos(|Z |1/2) if Z < 0cosh(Z 1/2) if Z ≥ 0

η0(Z ) =

sin(|Z |1/2)/|Z |1/2 if Z < 01 if Z = 0sinh(Z 1/2)/Z 1/2 if Z > 0

Z := (µ h)2 = −(ω h)2

ηn(Z ) :=1Z

[ηn−2(Z ) − (2n − 1)ηn−1(Z )], n = 1, 2, 3, . . .

η′n(Z ) =12ηn+1(Z ), n = 1, 2, 3, . . .

η−1(Z ) =

η0(Z ) =

Z := (µ h)2 = −(ω h)2

ηn(Z ) :=1Z

[ηn−2(Z ) − (2n − 1)ηn−1(Z )], n = 1, 2, 3, . . .

η′n(Z ) =12ηn+1(Z ), n = 1, 2, 3, . . .

η−1(Z ) =

η0(Z ) =

Z := (µ h)2 = −(ω h)2

ηn(Z ) :=1Z

[ηn−2(Z ) − (2n − 1)ηn−1(Z )], n = 1, 2, 3, . . .

η′n(Z ) =12ηn+1(Z ), n = 1, 2, 3, . . .

η−1(Z ) =

η0(Z ) =

Z := (µ h)2 = −(ω h)2

ηn(Z ) :=1Z

[ηn−2(Z ) − (2n − 1)ηn−1(Z )], n = 1, 2, 3, . . .

η′n(Z ) =12ηn+1(Z ), n = 1, 2, 3, . . .

Choice of ω

• local optimizationbased on local truncation error (lte)ω is step-dependent

• global optimizationPreservation of geometric properties (periodicity, energy,. . . )ω is constant over the interval of integration

Choice of ω

• local optimizationbased on local truncation error (lte)ω is step-dependent

• global optimizationPreservation of geometric properties (periodicity, energy,. . . )ω is constant over the interval of integration

Fourth-order boundary value problem

y (4) = F (t , y) a ≤ t ≤ b

y(a) = A1 y ′′(a) = A2

y(b) = B1 y ′′(b) = B2

• special case : y (4) + f (t) y = g(t)

• mathematical modeling of viscoelastic and inelastic flows,deformation of beams, plate deflection theory, . . .

• work by Doedel, Usmani, Agarwal, Cherruault et al., VanDaele et al., . . .

• finite differences, B-splines, . . .

y (4) = F (t , y) a ≤ t ≤ b

y(a) = A1 y ′′(a) = A2

y(b) = B1 y ′′(b) = B2

y (4) = F (t , y) a ≤ t ≤ b

y(a) = A1 y ′′(a) = A2

y(b) = B1 y ′′(b) = B2

y (4) = F (t , y) a ≤ t ≤ b

y(a) = A1 y ′′(a) = A2

y(b) = B1 y ′′(b) = B2

y (4) = F (t , y) a ≤ t ≤ b

y(a) = A1 y ′′(a) = A2

y(b) = B1 y ′′(b) = B2

The formulae

tj = a + j h, j = 0, 1, . . . N + 1 N ≥ 3 h :=b − aN + 1

• central formula for j = 2, . . . , N − 1

yj−2 + a1 yj−1 + a0 yj + a1 yj+1 + yj+2 =

b2 Fj−2 + b1 Fj−1 + b0 Fj + b1 Fj+1 + b2 Fj+2)

whereby yj is approximate value of y(tj) and Fj := F (tj , yj).

• begin formula

y0 + α1 y1 + α2 y2 + a3 y3 =

γ h2 y ′′0 + h4 (β0 F0 + β1 F1 + β2 F2 + β3 F3 + β4 F4 + β5 F5)

• end formula

The formulae

tj = a + j h, j = 0, 1, . . . N + 1 N ≥ 3 h :=b − aN + 1

yj−2 + a1 yj−1 + a0 yj + a1 yj+1 + yj+2 =

• begin formula

y0 + α1 y1 + α2 y2 + a3 y3 =

• end formula

The formulae

tj = a + j h, j = 0, 1, . . . N + 1 N ≥ 3 h :=b − aN + 1

yj−2 + a1 yj−1 + a0 yj + a1 yj+1 + yj+2 =

• begin formula

y0 + α1 y1 + α2 y2 + a3 y3 =

• end formula

The formulae

tj = a + j h, j = 0, 1, . . . N + 1 N ≥ 3 h :=b − aN + 1

yj−2 + a1 yj−1 + a0 yj + a1 yj+1 + yj+2 =

• begin formula

y0 + α1 y1 + α2 y2 + a3 y3 =

• end formula

Central formula

L[y ] := y(t − 2 h) + a1 y(t − h) + a0 y(t) + a1 y(t + h) + y(t + 2 h)

−h4(

b2 y (4)(t − 2 h) + b1 y (4)(t − h) + b0 y (4)(t) + b1 y (4)(t + h) + b2 y (4)(t + 2 h))

P = −1 : L[y ] = 0 for y ∈ S ={

1, t , t2, . . . , tM−1}

Central formula

−h4(

P = −1 : L[y ] = 0 for y ∈ S ={

1, t , t2, . . . , tM−1}

M = 10 :

yp−2 − 4 yp−1 + 6 yp − 4 yp+1 + yp+2 =

−y (4)p−2 + 124 y (4)

p−1 + 474 y (4)p + 124 y (4)

p+1 − y (4)p+2

L[y ](t) =1

3024h10 y (10)(t) + O(h12)

Central formula

−h4(

P = −1 : L[y ] = 0 for y ∈ S ={

1, t , t2, . . . , tM−1}

M = 8 and b2 = 0 :

yp−2 −4 yp−1 +6 yp −4 yp+1 +yp+2 =h4

y (4)p−1 + 4 y (4)

p + y (4)p+1

L[y ](t) = −1

720h8 y (8)(t) + O(h10)

Central formula

−h4(

P = −1 : L[y ] = 0 for y ∈ S ={

1, t , t2, . . . , tM−1}

M = 6 and b1 = b2 = 0 :

yp−2 − 4 yp−1 + 6 yp − 4 yp+1 + yp+2 = h4 y (4)p

L[y ](t) =16

h6 y (6)(t) + O(h8)

EF Central formula

−h4(

P = 0 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), 1, t , t2, . . . , tM−3}

EF Central formula

−h4(

P = 0 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), 1, t , t2, . . . , tM−3}

M = 10 :

yp−2 − 4 yp−1 + 6 yp − 4 yp+1 + yp+2 =

b2 y (4)p−2 + b1 y (4)

p−1 + b0 y (4)p + b1 y (4)

p+1 + b2 y (4)p+2

b0 =4 cos2 θ − 2 − 11 cos θ

6 (cos θ − 1)2+

θ4b1 =

cos2 θ + 5

6 (cos θ − 1)2−

θ4b2 = −

cos θ + 2

12 (cos θ − 1)2+

L[y ](t) =1

3024h10

y (10)(t) + ω2 y (8)(t))

+ O(h12)

EF Central formula

−h4(

P = 0 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), 1, t , t2, . . . , tM−3}

M = 8 and b2 = 0 :

yp−2−4 yp−1+6 yp−4 yp+1+yp+2 = h4(

b1 y (4)p−1 + b0 y (4)

p + b1 y (4)p+1

b0 =cos θ

cos θ − 1+

4 (1 − cos θ)

θ4b1 =

2 (1 − cos θ)+

2 (cos θ − 1)

L[y ](t) = −1

y (8)(t) + ω2 y (6)(t))

+ O(h10)

EF Central formula

−h4(

P = 0 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), 1, t , t2, . . . , tM−3}

M = 6 and b1 = b2 = 0 :

yp−2 − 4 yp−1 + 6 yp − 4 yp+1 + yp+2 =sin4(θ/2)

(θ/2)4 h4 y (4)p

L[y ](t) =16

h6 (y (6)(t) + ω2 y (4)(t)) + O(h8)

EF Central formula

−h4(

P = 1 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), t cos(ω t), t sin(ω t), 1, t , t2, . . . , tM−5}

EF Central formula

−h4(

P = 1 : L[y ] = 0 for y ∈ S ={

cos(ω t), sin(ω t), t cos(ω t), t sin(ω t), 1, t , t2, . . . , tM−5}

M = 6 and b1 = b2 = 0 :

yp−2 + a1 yp−1 + a0 yp + a1 yp+1 + yp+2 = b0 h4 y (4)p

a0 = 2−8 sin2 θ + θ (4 cos θ − 1) sin θ − 4 cos θ + 4

θ sin θ + 4 cos θ − 4a1 = −4

sin θ (θ cos θ − 2 sin θ)

θ sin θ + 4 cos θ − 4

b0 = 4sin θ

sin2 θ − 2 + 2 cos θ)

θ3 (θ sin θ + 4 cos θ − 4)

L[y ](t) =16

h6 (y (6)(t) + 2 ω2 y (4)(t) + ω4 y (2)(t)) + O(h8)

Coefficients of Central formula M = 6

0 2.0375 4.075−4

P=−1P=0P=1P=2

0 2.0375 4.075−14

P=−1P=0P=1P=2

0 2.0375 4.0750

P=−1P=0P=1P=2

0 1 25.8

P=−1P=0P=1P=2P=3

0 1 2−4.2

−3.8

P=−1P=0P=1P=2P=3

0 1 20.647

P=−1P=0P=1P=2P=3

0 1 20.147

P=−1P=0P=1P=2P=3

0 1 25.8

P=−1P=0P=1P=2P=3P=4

0 1 2−4.2

−3.8

P=−1P=0P=1P=2P=3P=4

0 1 20.639

0 1 20.153

0 1 2−0.021

−0.001

Central formula : coefficientsE.g. b0 in case M = 6In closed form . . .

• P = −1 :b0 = 1

• P = 0 :

b0 = 4(cos θ − 1)2

• P = 1 :

b0 = −4sin θ (cos θ − 1)2

θ3 (4 cos θ − 4 + θ sin θ)

• P = 2 :

b0 = −2sin3 θ

θ2 (θ cos θ − 3 sin θ)

Central formula : coefficientsE.g. b0 in case M = 6As a series . . .

• P = −1 :b0 = 1

• P = 0 :

b0 = 1 −16θ2 +

θ4 + O(θ6)

• P = 1 :

b0 = 1 −13θ2 +

θ4 + O(θ6)

• P = 2 :

b0 = 1 −12θ2 +

θ4 + O(θ6)

Central formula : local truncation error

lte = L[y ](t)

As an inifinite series :

lte = hM CM DK+1 (D2 + ω2)P+1y(t) + O(hM+2)

In closed form : (Coleman and Ixaru)

lte = hM ΦK ,P(Z ) DK+1 (D2 + ω2)P+1y(ξ)

Z ∈ some interval ΦK ,P(0) 6= 0 ξ ∈ (t − 2 h, t + 2 h)

lte = L[y ](t)

Local truncation error

lte = hM CM DK+1 (D2 + ω2)P+1y(t) + O(hM+2) ,

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

• P = 0 :y (K+3)(tj) + y (K+1)(tj)ω2

• P = 1 :

y (K+5)(tj) + 2 y (K+3)(tj)ω2j + y (K+1)(tj)ω4

• P = 2 :

y (K+7)(tj)+3 y (K+5)(tj)ω4j +3 y (K+3)(tj)ω4

j +y (K+1)(tj)ω6j = 0

• . . .

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

• P = 0 :y (K+3)(tj) + y (K+1)(tj)ω2

• P = 1 :

• P = 2 :

j +y (K+1)(tj)ω6j = 0

• . . .

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

• P = 0 :y (K+3)(tj) + y (K+1)(tj)ω2

• P = 1 :

• P = 2 :

j +y (K+1)(tj)ω6j = 0

• . . .

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

• P = 0 :y (K+3)(tj) + y (K+1)(tj)ω2

• P = 1 :

• P = 2 :

j +y (K+1)(tj)ω6j = 0

• . . .

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

• P = 0 :y (K+3)(tj) + y (K+1)(tj)ω2

• P = 1 :

• P = 2 :

j +y (K+1)(tj)ω6j = 0

• . . .

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

ω2j is solution of equation of degree P + 1.

• Which value of P should be chosen ?

• Which root ωj should be chosen ?

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

At tj : D(K+1) (D2 + ω2j )(P+1)y(t)

t=tj= 0 j = 2, . . . , N − 1

Parameter selection

lte = hM CM DK+1 (D2 − µ2)P+1y(t) + O(hM−2)

Suppose y(t) takes the form tP0 eµ0 t

Then lte= 0 for any EF rule with P ≥ P0 and µj = µ0

TheoremIf y(t) = tP0 eµ0 t then ν = µ2

0 is a root of multiplicity P − P0 + 1of DK+1 (D2 − ν)P+1y(t) = 0.

• if P = P0, then µ = µ0 will be a single root

• if P = P0 + 1, then µ = µ0 will be a double root

• if P = P0 + 2, then µ = µ0 will be a triple root

• . . .

Parameter selection

• . . .

Parameter selection

• . . .

Parameter selection

• . . .

Parameter selection

Suppose y(t) does not take the form tP0 eµ0 t

Then y(t) 6∈ S for any P.

For a given value of P :

D(K+1) (D2 − µ2j )

(P+1)y(t)∣

t=tj= 0

At each point tj , this gives P + 1 values for µ2j .

Idea : keep |µj h| as small as possible.If possible, choose P ≥ 1 to avoid too large values for |µj |.

Parameter selection

D(K+1) (D2 − µ2j )

(P+1)y(t)∣

t=tj= 0

Parameter selection

D(K+1) (D2 − µ2j )

(P+1)y(t)∣

t=tj= 0

Parameter selection

D(K+1) (D2 − µ2j )

(P+1)y(t)∣

t=tj= 0

Parameter selection

D(K+1) (D2 − µ2j )

(P+1)y(t)∣

t=tj= 0

First example

y (4) −384 t4

(2 + t2)4 y = 242 − 11 t2

(2 + t2)4

y(−1) =13

y(1) =13

y ′′(−1) =2

27y ′′(1) =

Solution : y(t) =1

2 + t2

First example

y (4) −384 t4

(2 + t2)4 y = 242 − 11 t2

(2 + t2)4

y(−1) =13

y(1) =13

y ′′(−1) =2

27y ′′(1) =

Solution : y(t) =1

2 + t2

µj for M = 8

P = 0 : y (8)(tj) − y (6)(tj)µ2j = 0

• re-express higher orderderivatives in terms of y ,y ′, y ′′ and y ′′′

• approximate y ′, y ′′ andy ′′′ in terms of y

• an initial approximationfor y can be computedwith a polynomial rule

µj for M = 8

P = 0 : y (8)(tj) − y (6)(tj)µ2j = 0

µj for M = 8

P = 0 : y (8)(tj) − y (6)(tj)µ2j = 0

µj for M = 8

P = 0 : y (8)(tj) − y (6)(tj)µ2j = 0

µj for M = 8

P = 0 : y (8)(tj) − y (6)(tj)µ2j = 0

−1 −0.5 0 0.5 1

Real and imaginary partof µj

µj for M = 8

P = 1 : y (8)(tj) − 2 y (6)(tj)µ2j + y (4)(tj)µ4

−1 −0,5 0 0,5 1−5

Real and imag. part of µ1,j and µ2,j

µj for M = 8

P = 1 : y (8)(tj) − 2 y (6)(tj)µ2j + y (4)(tj)µ4

−1 −0,5 0 0,5 1−5

Real and imag. part of µ1,j and µ2,j

−1 −0,5 0 0,5 1

Real and imaginarypart of µj with smallestnorm

µj for M = 8

P = 1 : y (8)(tj) − 2 y (6)(tj)µ2j + y (4)(tj)µ4

10−3

10−2

10−1

10−10

10−9

10−8

10−7

10−6

error obtained with µ1,j , µ2,j and µ with smallest norm

Global error

M = 6 :(K , P) = (5,−1) : second-order method(K , P) = (1, 1) : fourth-order method

10−3

10−2

10−1

10−10

10−8

10−6

10−4

10−2

Global error

M = 8 :(K , P) = (7,−1) : fourth-order method(K , P) = (3, 1) : sixth-order method

10−3

10−2

10−1

10−12

10−10

10−8

10−6

10−4

Global error

M = 10 :(K , P) = (9,−1) : sixth-order method(K , P) = (5, 1) : eighth-order method

10−3

10−2

10−1

10−12

10−10

10−8

10−6

Condition number

10−3

10−2

10−1

Second example

y (4) − t = 4 et

y(−1) = −1/e y(1) = e

y ′′(−1) = 1/e y ′′(1) = 3 e

Solution : y(t) = et t

Second example

y (4) − t = 4 et

y(−1) = −1/e y(1) = e

y ′′(−1) = 1/e y ′′(1) = 3 e

Solution : y(t) = et t

µj for M = 6

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

differentiating the differential equation :

(y (2)(tj) + 4 etj ) − 2 (yj + 4 etj )µ2j + y (2)(tj)µ4

y (2)(tj) approximated by fourth-order finite difference scheme

µj for M = 6

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

µj for M = 6

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

µj for M = 6

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

−1 −0.5 0 0.5 10

two real roots µ(1) and µ(2)

µj for M = 6

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

−1 −0.5 0 0.5 10

two real roots µ(1) and µ(2)10

−310

−210

10−8

10−7

10−6

10−5

10−4

10−3

10−2

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

y (2)(tj) approximated by sixth-order finite difference scheme

−1 −0.5 0 0.5 10

two real roots µ(1) and µ(2)

P = 1 : y (6)(tj) − 2 y (4)(tj)µ2j + y (2)(tj)µ4

y (2)(tj) approximated by sixth-order finite difference scheme

−1 −0.5 0 0.5 10

two real roots µ(1) and µ(2)10

−310

−210

−110

10−12

10−10

10−8

10−6

10−4

10−2

Conclusions

• Fourth-order boundary value problems are solved bymeans of parameterized exponentially-fitted methods.

• A suitable value for the parameter can be found from theroots of the leading term of the local truncation error.

• If a constant value is found, then a very accurate solutioncan be obtained.

• However, the methods strongly suffer from the fact that thesystem to be solved is ill-conditioned for small values of themesh size.

Conclusions

exponentially-fitted methods applied to fourth-order boundary ...mvdaele/voordrachten/kos2008.pdf•...

Documents

marnix.vandaele@ugentmvdaele/voordrachten/zaragoza... ·...

7. analytical methods well-established methods that are used...

jaargang 17 / 2011 amsterdam economic board: het gaat nu...

congres persoonsgebonden budget - guus...

natuurkundige voordrachten 2015 - 2016 - … · 2020. 5....

waar gebeurt het in fryslân? - university of groningen ›...

qualitative methods vs. quantitative methods. qualitative...

· web viewherzieningswet artikel 30.9 regelt het aantal...

analytical methods - algebraic methods

chapter 5- even more about objects and methods. overview n...

itgebreid cv (inclusief voordrachten en andere...

numerical methods specific methods: finite differences...

erkenningsnummer p508195 wijkteams kortrijk het oekske ·...

samenvattingen van de voordrachten van 23 januari 2021

hoofdstuk 10 -...

uitgeverij verloren filepr passage reeks sh serta historica....

i -...

materials and methods: computational methods

voltammetric methods and electrodes. electroanalytical...

muziek, theater, voordrachten, wandelingen, kunstprojecten...